基于深度网络的滚动轴承智能故障诊断

李金才; 付文龙; 王仁明; 陈星; 孟嘉鑫

doi:10.13272/j.issn.1671-251x.2022010008

基于深度网络的滚动轴承智能故障诊断

doi: 10.13272/j.issn.1671-251x.2022010008

李金才^1,,
付文龙^{1, 2, 3, ,},
王仁明¹,
陈星¹,
孟嘉鑫¹

1.
三峡大学电气与新能源学院, 湖北宜昌　443002
2.
三峡大学梯级水电站运行与控制湖北省重点实验室, 湖北宜昌　443002
3.
三峡大学水电动机械设备设计与维护湖北省重点实验室, 湖北宜昌　443002

基金项目: 国家自然科学基金资助项目(51741907)；湖北省水电动机械设计与维修重点实验室开放基金（2020KJX03）。

详细信息

作者简介:
李金才（1996-），男，河南开封人，硕士研究生，主要研究方向为机械故障诊断与人工智能，E-mail:2953700543@qq.com

通讯作者:
付文龙（1988-），男，湖北仙桃人，副教授，博士，主要研究方向为信号处理与旋转机械系统状态分析，E-mail：ctgu_fuwenlong@126.com

中图分类号: TD712
计量
- 文章访问数: 554
- HTML全文浏览量: 147
- PDF下载量: 46
- 被引次数: 0
出版历程
- 收稿日期: 2022-01-05
- 修回日期: 2022-04-05
- 网络出版日期: 2022-04-08

Intelligent fault diagnosis of rolling bearings based on deep network

LI Jincai^1
,,
FU Wenlong^{1, 2, 3
, ,},
WANG Renming¹,
CHEN Xing¹,
MENG Jiaxin¹

1.
College of Electrical Engineering & New Energy, China Three Gorges University, Yichang 443002, China
2.
Hubei Key Laboratory of Cascaded Hydropower Stations Operation & Control, China Three Gorges University, Yichang 443002, China
3.
Hubei Key Laboratory of Hydropower Machinery Design and Maintenance, China Three Gorges University, Yichang 443002, China

摘要

摘要: 针对变工况环境中滚动轴承的源域与目标域数据分布不同及目标域样本不含标签的问题，提出一种基于深度自适应迁移学习网络(DATLN)的滚动轴承故障诊断模型。首先，搭建领域共享特征提取网络，采用多尺度卷积神经网络(MSCNN)抑制噪声的干扰，进而有效提取振动信号中蕴含的局部故障信息；其次，结合双向长短时记忆网络(BiLSTM)进一步学习局部故障信息中的时间特征；最后，引入迁移学习，以域对抗(DA)训练结合自适应联合分布(AJD)度量构建域自适应模块，通过最大化域分类损失和最小化AJD距离，实现源域与目标域特征样本对齐。在开源CWRU数据集与机械故障平台实测数据集上分别进行抗噪实验和迁移实验。抗噪实验表明：① 在无噪声环境下，MSCNN−BiLSTM网络的识别准确率均达到99%以上，说明其具有较好的特征提取能力；② MSCNN−BiLSTM，LeNet−5，MSCNN和BiLSTM四种网络的识别准确率随着噪声强度的增强而降低；③ 在3，5，10 dB噪声环境下，MSCNN−BiLSTM网络的平均识别准确率比LeNet−5，MSCNN和BiLSTM 网络的平均识别准确率均高，说明MSCNN−BiLSTM网络具有较好的抗噪声干扰性能；④ MSCNN−BiLSTM网络在无噪声环境和3 dB噪声环境下，均最先达到收敛且波动较小。迁移实验表明：① 在无标签目标域数据集上，DA+AJD方法的平均识别准确率为97.36%，均高于Baseline，迁移成分分析(TCA)，域对抗神经网络（DANN）的识别准确率；② 在测试集混淆矩阵上，DA+AJD方法仅有1个样本被错误识别，表明基于域适应的DA+AJD方法具备更好的故障迁移诊断性能；③ 利用t−SNE算法对处理后的源域与目标域特征样本进行可视化，DA+AJD方法只有少量目标域的滚动体故障和外圈故障特征样本被错误对齐到源域的内圈故障特征样本区域，说明DA+AJD方法可有效减少源域与目标域的边缘分布和条件分布差异，进而达到更好的特征样本对齐效果。
- 滚动轴承 /
- 智能故障诊断 /
- 多尺度卷积神经网络 /
- 无标签目标域样本 /
- 深度学习 /
- 迁移学习 /
- 自适应联合分布
Abstract: In order to solve the problem that the data distribution of the source domain and the target domain of rolling bearing is different in the variable working condition environment and the samples of the target domain do not contain labels, a fault diagnosis model of the rolling bearing based on the deep adaptive transfer learning network (DATLN) is proposed. Firstly, a domain-shared characteristic extraction network is built, and multiscale convolutional neural network (MSCNN) is used to suppress noise interference, so as to effectively extract local fault information contained in vibration signals. Secondly, combined with a bi-directional long short-term memory network (BiLSTM), the temporal characteristics in the local fault information are further learned. Finally, transfer learning is introduced to build a domain adaptive module with domain adversarial (DA) training combined with adaptive joint distribution (AJD) metrics. By maximizing the domain classification loss and minimizing the AJD distance, the source and target domain characteristic samples are aligned. The anti-noise experiment and transfer experiment are carried out on the open source CWRU data set and the measured data set of the mechanical fault platform respectively. The anti-noise experiments show the following points. ① The identification accuracy of MSCNN-BiLSTM network is above 99% in the noise-free environment, which shows that MSCNN-BiLSTM network has a good characteristic extraction capability. ② The identification accuracy of MSCNN-BiLSTM, LeNet-5, MSCNN and BiLSTM decreases with the increase of noise intensity. ③ Under the noise environment of 3, 5 and 10 dB, the average identification accuracy of MSCNN-BiLSTM network is higher than that of LeNet-5, MSCNN and BiLSTM networks, indicating that MSCNN-BiLSTM network has better anti-noise interference performance. ④ The MSCNN-BiLSTM network converges first with less fluctuation in both the noise-free environment and the 3 dB noise environment. The transfer experiments show the following points. ① The average identification accuracy of DA+AJD method is 97.36% on unlabeled target domain dataset, which is higher than that of Baseline, transfer component analysis(TCA) and domain adversarial neural network (DANN). ② On the test set confusion matrix, only one sample of the DA+AJD method is incorrectly identified, indicating that the DA+AJD method based on domain adaptation has better fault transfer diagnosis performance. ③ The t-SNE algorithm is used to visualize the processed source and target domain characteristic samples. The DA+AJD method only has a small number of rolling element fault and outer ring fault characteristic samples in the target domain that are incorrectly aligned to the inner ring fault characteristic samples area in the source domain. This result indicates that the DA+AJD method can effectively reduce the edge distribution and conditional distribution differences between the source domain and the target domain, and thus achieves better characteristic sample alignment.
- rolling bearing /
- intelligent fault diagnosis /
- multiscale convolutional neural network /
- unlabeled target domain sample /
- deep learning /
- transfer learning /
- adaptive joint distribution

HTML全文

图 1 迁移学习

Figure 1. Transfer learning

下载: 全尺寸图片幻灯片

图 2 BiLSTM网络结构

Figure 2. Structure of BiLSTM network

下载: 全尺寸图片幻灯片

图 3 MSCNN−BiLSTM网络

Figure 3. MSCNN-BilSTM network

下载: 全尺寸图片幻灯片

图 4 滚动轴承故障诊断模型

Figure 4. Model of rolling bearing fault diagnosis

下载: 全尺寸图片幻灯片

图 5 CWRU轴承数据采集系统

Figure 5. CWRU bearing data acquisition system

下载: 全尺寸图片幻灯片

图 6 不重叠采样

Figure 6. Non-overlapping sampling

下载: 全尺寸图片幻灯片

图 7 正常状态下振动信号变化

Figure 7. Vibration signal changes under the normal state

下载: 全尺寸图片幻灯片

图 8 内圈故障（IR07）状态下振动信号变化

Figure 8. Vibration signal changes in the inner fault （IR07） state

下载: 全尺寸图片幻灯片

图 9 无噪声环境下对比实验结果

Figure 9. Comparison of experimental results in noiseless environment

下载: 全尺寸图片幻灯片

图 12 3 dB噪声环境下对比实验结果

Figure 12. Comparison of experimental results in 3 dB environment

下载: 全尺寸图片幻灯片

图 11 5 dB噪声环境下对比实验结果

Figure 11. Comparison of experimental results in 5 dB environment

下载: 全尺寸图片幻灯片

图 10 10 dB噪声环境下对比实验结果

Figure 10. Comparison of experimental results in 10 dB environment

下载: 全尺寸图片幻灯片

图 13 无噪声环境下0负载测试集识别结果

Figure 13. Identification results of 0 load test set in noise-free environment

下载: 全尺寸图片幻灯片

图 14 3 dB噪声环境下0负载测试集识别结果

Figure 14. Identification results of 0 load test set in 3 dB environment

下载: 全尺寸图片幻灯片

图 15 机械故障模拟实验台

Figure 15. Machinery fault simulator

下载: 全尺寸图片幻灯片

图 16 3 dB噪声环境下迁移结果

Figure 16. Transfer results of 3 dB environment

下载: 全尺寸图片幻灯片

图 17 迁移任务C to B的测试集混淆矩阵

Figure 17. Test dataset confusion matrix of transfer task C to B

下载: 全尺寸图片幻灯片

图 18 迁移任务C to B的t-SNE特征可视化

Figure 18. T-SNE characteristic visualization of transfer task C to B

下载: 全尺寸图片幻灯片

表 1 MSCNN−BiLSTM网络参数

Table 1. Parameters of MSCNN-BiLSTM network

网络	层类型	核尺寸/步长	核数量	激活函数	输入尺寸	输出尺寸
MSCNN网络通道1	卷积层1	15/1	16	ReLU	（1，1 024）	（16，1 010）
	卷积层2	15/1	32	ReLU	（16，1 010）	（32，996）
	最大池化层	2/2	—	—	（32，996）	（32，498）
	卷积层3	15/1	64	ReLU	（32，498）	（64，484）
	卷积层4	15/1	128	ReLU	（64，484）	（128，470）
	自适应最大池化层	—	—	—	（128，470）	（128，4）
MSCNN网络通道2	卷积层5	5/1	16	ReLU	（1，1 024）	（16，1 020）
	卷积层6	5/1	32	ReLU	（16，1 020）	（32，1 016）
	最大池化层	2/2	—	—	（32，1 016）	（32，508）
	卷积层7	5/1	64	ReLU	（32，508）	（64，504）
	卷积层8	5/1	128	ReLU	（64，504）	（128，500）
	自适应最大池化层	—	—	—	（128，500）	（128，4）
MSCNN网络汇聚层	汇聚层	—	—	—	（128，4），（128，4）	（128，4）
BiLSTM网络	BiLSTM层	—	—	ReLU	（128，4）	（256）

下载: 导出CSV

表 2 域分类器参数

Table 2. Parameters of domain classifier

层次	神经元个数
全连接层1	256
全连接层2	128
全连接层3	2

下载: 导出CSV

表 3 0负载下数据集

Table 3. Date set under 0 load

损伤直径/mm	损伤位置	标记
−	正常	N
0.177 8	内圈	IR07
0.355 6	内圈	IR14
0.533 4	内圈	IR21
0.177 8	外圈	OR07
0.355 6	外圈	OR14
0.533 4	外圈	OR21
0.177 8	滚动体	B07
0.355 6	滚动体	B14
0.533 4	滚动体	B21

下载: 导出CSV

表 4 CWRU样本集

Table 4. CWRU sample set

状态	标签	样本数
状态	标签	0	0.75 kW	1.5 kW	2.25 kW
N	0	100	100	100	100
IR07	1	100	100	100	100
B07	2	100	100	100	100
OR07	3	100	100	100	100
IR14	4	100	100	100	100
B14	5	100	100	100	100
OR14	6	100	100	100	100
IR21	7	100	100	100	100
B21	8	100	100	100	100
OR21	9	100	100	100	100

下载: 导出CSV

表 5 不同网络的平均识别准确率

Table 5. Average accuracy of different network

网络	平均识别准确率/%
网络	3 dB	5 dB	10 dB
LeNet−5	90.74	93.83	95.42
MSCNN	95.57	96.89	97.14
BiLSTM	89.10	92.58	96.99
MSCNN−BiLSTM	98.43	99.00	99.16

下载: 导出CSV

表 6 每种方法的平均识别准确率

Table 6. Average results of different methods

方法	平均识别准确率/%
Baseline	75.90
TCA	85.38
DANN	87.19
DA+AJD	97.36

下载: 导出CSV

参考文献(17)

[1]	宫涛,杨建华,单振,等. 强噪声背景与变转速工况条件下滚动轴承故障诊断研究[J]. 工矿自动化,2021,47(7):63-71. GONG Tao,YANG Jianhua,SHAN Zhen,et al. Research on rolling bearing fault diagnosis under strong noise background and variable speed working condition[J]. Industry and Mine Automation,2021,47(7):63-71.
[2]	鞠晨,张超,樊红卫,等. 基于小波包分解和PSO-BPNN的滚动轴承故障诊断[J]. 工矿自动化,2020,46(8):70-74. JU Chen,ZHANG Chao,FAN Hongwei,et al. Rolling bearing fault diagnosis based on wavelet packet decomposition and PSO-BPNN[J]. Industry and Mine Automation,2020,46(8):70-74.
[3]	LI Xiang,ZHANG Wei,DING Qian,et al. Multi-layer domain adaptation method for rolling bearing fault diagnosis[J]. Signal Processing,2019,157:180-197. doi: 10.1016/j.sigpro.2018.12.005
[4]	WANG Yujing, WANG Chao, KANG Shouqiang, et al. The network combined broad learning with transfer learning: a new intelligent fault diagnosis method for rolling bearings[J]. Measurement Science and Technology, 2020, 31（11）. DOI: 10.1088/1361-6501/ab8fee.
[5]	CHEN Zhuyun,GRYLLIAS K,LI Weihua. Intelligent fault diagnosis for rotary machinery using transferable convolutional neural network[J]. IEEE Transactions on Industrial Informatics,2019,16(1):339-349.
[6]	陈超,沈飞,严如强. 改进LSSVM迁移学习方法的轴承故障诊断[J]. 仪器仪表学报,2017,38(1):33-40. doi: 10.3969/j.issn.0254-3087.2017.01.005 CHEN Chao,SHEN Fei,YAN Ruqiang. Enhanced least squares support vector machine-based transfer learning strategy for bearing fault diagnosis[J]. Chinese Journal of Scientific Instrument,2017,38(1):33-40. doi: 10.3969/j.issn.0254-3087.2017.01.005
[7]	陈仁祥,陈思杨,杨黎霞,等. 改进TrAdaBoost多分类算法的滚动轴承故障诊断[J]. 振动与冲击,2019,38(15):36-41,48. CHEN Renxiang,CHEN Siyang,YANG Lixia,et al. Fault diagnosis of rolling bearings based on improved TrAdaBoost multi-classification algorithm[J]. Journal of Vibration and Shock,2019,38(15):36-41,48.
[8]	康守强,邹佳悦,王玉静,等. 基于无监督特征对齐的变负载下滚动轴承故障诊断方法[J]. 中国电动机工程学报,2020,40(1):274-281,393. KANG Shouqiang,ZOU Jiayue,WANG Yujing,et al. Fault diagnosis method of a rolling bearing under varying loads based on unsupervised feature alignment[J]. Proceedings of the CSEE,2020,40(1):274-281,393.
[9]	GUO Liang,LEI Yaguo,XING Saibo,et al. Deep convolutional transfer learning network:a new method for intelligent fault diagnosis of machines with unlabeled data[J]. IEEE Transactions on Industrial Electronics,2018,66(9):7316-7325.
[10]	AN Zenghui,LI Sunming,WANG Jinrui,et al. Generalization of deep neural network for bearing fault diagnosis under different working conditions using multiple kernel method[J]. Neurocomputing,2019,352(8):42-53.
[11]	HAN Te,LIU Chao,YANG Wenguang,et al. A novel adversarial learning framework in deep convolutional neural network for intelligent diagnosis of mechanical faults[J]. Knowledge-Based Systems,2019,165(2):474-487.
[12]	CHENG Cheng,ZHOU Beitong,MA Guijun,et al. Wasserstein distance based deep adversarial transfer learning for intelligent fault diagnosis with unlabeled or insufficient labeled data[J]. Neurocomputing,2020,409:35-45. doi: 10.1016/j.neucom.2020.05.040
[13]	吴静然,刘建华,崔冉. 子域适应无监督轴承故障诊断[J]. 振动与冲击,2021,40(15):34-40. WU Jingran,LIU Jianhua,CUI Ran. Sub-domain adaptive unsupervised bearing fault diagnosis[J]. Journal of Vibration and Shock,2021,40(15):34-40.
[14]	许子非,金江涛,李春. 基于多尺度卷积神经网络的滚动轴承故障诊断方法[J]. 振动与冲击,2021,40(18):212-220. XU Zifei,JIN Jiangtao,LI Chun. New method for the fault diagnosis of rolling bearings based on a multiscale convolutional neural network[J]. Journal of Vibration and Shock,2021,40(18):212-220.
[15]	CHEN Xiaohan,ZHANG Beike,GAO Dong. Bearing fault diagnosis base on multi-scale CNN and LSTM model[J]. Journal of Intelligent Manufacturing,2021,32(4):971-987. doi: 10.1007/s10845-020-01600-2
[16]	XIE Junyao, ZHANG Laibin, DUAN Lixiang, et al. On cross-domain feature fusion in gearbox fault diagnosis under various operating conditions based on transfer component analysis[C]//2016 IEEE International Conference on Prognostics and Health Management （ICPHM）, Ottawa, 2016: 20-22.
[17]	HAN Te,LIU Chao,YANG Wenguang,et al. Deep transfer network with joint distribution adaptation:a new intelligent fault diagnosis framework for industry application[J]. ISA Transactions,2020,97(2):269-281.